by Bruce Schneier
Founder and CTO
Counterpane Internet Security, Inc.
Copyright (c) 2000 by Counterpane Internet Security, Inc.
The goal of any voting system is to establish the intent of the voter, and transfer that intent to the vote counter. Amongst a circle of friends, a show of hands can easily decide which movie to attend. The vote is open and everyone can monitor it. But what if Alice wants _Charlie's Angels_ and Bob wants _102 Dalmatians_? Will Alice vote in front of his friends? Will Bob? What if the circle of friends is two hundred; how long will it take to count the votes? Will the theater still be showing the movie? Because the scale changes, our voting methods have to change.
Anonymity requires a secret ballot. Scaling and speed requirements lead to mechanical and computerized voting systems. The ideal voting technology would have these five attributes: anonymity, scalability, speed, audit, and accuracy -- direct mapping from intent to counted vote.
Through the centuries, different technologies have done their best. Stones and pot shards dropped in Greek vases led to paper ballots dropped in sealed boxes. Mechanical voting booths and punch cards replaced paper ballots for faster counting. New computerized voting machines promise even more efficiency, and Internet voting even more convenience.
But in the rush to improve the first four attributes, accuracy has been sacrificed. The way I see it, all of these technologies involve translating the voter's intent in some way; some of them involve multiple translations. And at each translation step, errors accumulate.
This is an important concept, and one worth restating. Accuracy is not how well the ballots are counted by, for example, the optical scanner; it's how well the process translates voter intent into properly counted votes.
Most of Florida's voting irregularities are a direct result of these translation errors. The Palm Beach system had several translation steps: voter to ballot to punch card to card reader to vote tabulator to centralized total. Some voters were confused by the layout of the ballot, and mistakenly voted for someone else. Others didn't punch their ballots so that the tabulating machines could read them. Ballots were lost and not counted. Machines broke down, and they counted ballots improperly. Subtotals were lost and not counted in the final total.
Certainly Florida's antiquated voting technology is partially to blame, but newer technology wouldn't magically make the problems go away. It could even make things worse, by adding more translation layers between the voters and the vote counters and preventing recounts.
That's my primary concern about computer voting: There is no paper ballot to fall back on. Computerized voting machines, whether they have keyboard and screen or a touch screen ATM-like interface, could easily make things worse. You have to trust the computer to record the votes properly, tabulate the votes properly, and keep accurate records. You can't go back to the paper ballots and try to figure out what the voter wanted to do. And computers are fallible; some of the computer voting machines in this election failed mysteriously and irrecoverably.
Online voting schemes have even more potential for failure and abuse. We know we can't protect Internet computers from viruses and worms, and that all the operating systems are vulnerable to attack. What recourse is there if the voting system is hacked, or simply gets overloaded and fails? There would be no means of recovery, no way to do a recount. Imagine if someone hacked the vote in Florida; redoing the election would be the only possible solution. A secure Internet voting system is theoretically possible, but it would be the first secure networked application ever created in the history of computers.
There are other, less serious, problems with online voting. First, the privacy of the voting booth cannot be imitated online. Second, in any system where the voter is not present, the ballot must be delivered tagged in some unique way so that people know it comes from a registered voter who has not voted before. Remote authentication is something we've not gotten right yet. (And no, biometrics don't solve this problem.) These problems also exist in absentee ballots and mail-in elections, and many states have decided that the increased voter participation is more than worth the risks. But because online systems have a central point to attack, the risks are greater.
The ideal voting system would minimize the number of translation steps, and make those remaining as simple as possible. My suggestion is an ATM-style computer voting machine, but one that also prints out a paper ballot. The voter checks the paper ballot for accuracy, and then drops it into a sealed ballot box. The paper ballots are the "official" votes and can be used for recounts, and the computer provides a quick initial tally.
Even this system is not as easy to design and implement as it sounds. The computer would need to be treated like safety- and mission-critical systems: fault tolerant, redundant, carefully analyzed code. Adding the printer adds problems; it's yet another part to fail. And these machines will only be used once a year, making it even harder to get right.
But in theory, this could work. It would rely on computer software, with all those associated risks, but the paper ballots would provide the ability to recount by hand if necessary.
Even with a system like this, we need to realize that the risk of errors and fraud cannot be brought down to zero. Cambridge Professor Roger Needham once described automation as replacing what works with something that almost works, but is faster and cheaper. We need to decide what's more important, and what tradeoffs we're willing to make.
This is *the* Web site on electronic voting. Rebecca Mercuri wrote her
PhD thesis on the topic, and it is well worth reading.
Good balanced essays:
Pro-computer and Internet voting essays:
Problems with New Mexico computerized vote-counting software: